There looks to be a slight concurrency problem in ReplicationQueue in the flush method (in HEAD).

 cache.callRemoteMethods(null, // send to all live nodes in the cluster
 TreeCache.replicateAllMethod, new Object[]{l}, false, true, 5000);
 elements.clear();

The elements.clear call is not atomic and another thread calling add between the callRemoteMethods and the clear would subsequently clear it.

I would suggest you move the clear up in the synchronized block after the clone.

 public void flush() {
 List l;
 synchronized(elements) {
 if(log.isInfoEnabled())
 log.info("flush(): flushing repl queue (num elements=" + elements.size() + ")");
 l=(List)elements.clone();
 elements.clear();
 }

 try {
 cache.callRemoteMethods(null, // send to all live nodes in the cluster
 TreeCache.replicateAllMethod, new Object[]{l}, false, true, 5000);
 }
 catch(Throwable t) {
 log.error("failed replicating " + l.size() + " elements in replication queue", t);
 }
 }